Why AI pilots stall
A skilled facilitator can make AI interesting to almost anyone. People leave curious, the evaluations glow, and leadership asks to roll it out to everyone.
But enjoying AI isn't using it. If people leave without finding an application for their own work - and without starting to build it - nothing changes on Monday.
Scale that up and you get more of the same. Reach goes up, relevance goes down, and attendance is the only number left to report.
Over the past few months we worked with a large insurer to do it differently. Using a lab approach, we combined hands-on learning with creating real AI tools (for this client, building Microsoft Copilot agents). Every participant leaves not only with working knowledge of AI basics, but with at least one AI-enabled tool they built for themselves.
After their labs, participants wrote short reflections on what they'd done with AI. More than 90% of leaders described a tool - an agent, a new workflow - they had already put to work. On the frontline, it was 98%.
Build real work in the room
Someone who leaves with a half-built tool for their own work has a reason to finish it. Someone who leaves with a certificate doesn't.
In LaunchPad, every participant builds a "ThinkFirst" agent - one that interviews them in depth about their work - then uses it to find use cases for AI in their own role. ThinkFirst then helps them conceptualize and build their own custom Copilot agent.
Executives in Ignite do the same in a 90-minute agent-building lab. Each one brings a use case from a process they own, builds a lighter version of ThinkFirst, and uses it to sharpen the idea. They leave with a working Copilot agent prototype. The commitment: build two agents before the next session. In the labs, every executive reported building their agents successfully.
Short daily follow-ups kept the learning and building going, because the gap between a good session and a changed habit takes weeks of practice, not one session.
One model, built for each job
Scale is where most programs break. A session trying to engage a senior executive, a claims specialist and a service rep works well for none of them.
So we built one model of good AI practice and adapted the lab to each audience:
- Leaders: LaunchPad, in-person and virtual labs.
- Executives: Ignite, 90-minute agent-building labs built around processes they own.
- Frontline teams: 15-minute online labs, with role-specific versions for five frontline teams, built with 1st90.
- Everyone: ten minutes a day of follow-up after each lab.
The frontline teams matter most. They handle a steady stream of customers and can't step away for a workshop, so most AI programs never reach them. Here, nobody had to leave the floor for two days. Nobody sat through examples from someone else's job.
Proof they kept building
We counted actions, not attendance. After each lab, participants commit to specific AI actions in their own work, then write a short reflection on what happened - all tracked in 1st90's app. Across leaders and frontline teams:
- Nearly 1,000 people enrolled: more than 500 leaders and 400 frontline colleagues.
- They've committed to more than 3,100 AI actions in their real work.
- 96% of written reflections describe a change already made - not a plan.
- 67% of frontline participants who started finished the full 11-step path. Eleven steps is a lot to ask of a working adult with a queue.
The same people rated themselves at the start and again at the end:
| Self-rating | Before | After |
|---|---|---|
| Leaders: spot where AI helps | 44% | 78% |
| Leaders: prompt with confidence | 51% | 69% |
| Leaders: use AI daily | 51% | 69% |
| Leaders: clear on safe use | 75% | 87% |
| Frontline: confident using AI | 27% | 56% |
| Frontline: clear on when to trust AI | 45% | 78% |
| Frontline: use AI most days | 36% | 52% |
Spotting where AI helps is judgment, the hardest part to teach, and it rose 34 points. Frontline confidence more than doubled, and knowing when to trust AI output jumped 33 points - exactly what you want from people who talk to customers all day.
And they kept building. One leader built an agent that consolidates two monthly reports, assigns cost centers, flags errors and hands Payroll a clean file. Another built two agents: a brainstorm partner, and a challenger whose only job is to poke holes in their ideas. On the frontline, people use AI to turn a few bullet points into a professional, personable customer message, compare two spreadsheets to find what's missing, and take a second look at case details a first review might have missed. One participant co-hosted an AI workshop for interns a few days after their own session.
What this means for your rollout
- Make every session a lab. People should leave with something they've started building for their own job.
- Build for each job, on one shared model. Different formats, same core.
- Measure what people build, not who showed up.
Hi is built for this kind of work. We design the labs and write every version of the content. Our partners bring what no single firm can: Advantage Performance Group administers the program end to end, and 1st90's platform puts learning inside the workday and tracks what people do next. That combination is how one model reaches every role - and how we can show what changed.
Ninety days from now, your people could be showing you what they built with AI - not just telling you they enjoyed the training. If that's the rollout you want, let's talk.
